Discovering Crystal Structure Prediction Algorithms with an AI Co-Scientist

2026-06-22

AI Summary

This work addresses the crystal structure prediction (CSP) problem by introducing HACO, a human-AI collaborative framework that pioneers the adaptation of the vision-based MaskGIT model to CSP. The proposed approach formulates a discretized, tokenized Masked Generative Crystal Transformer (MaskGXT), integrating crystallographic prior knowledge through symmetry-aware tokens, space-group hierarchical sampling, and sub-interval coordinate refinement. These mechanisms enforce domain-specific constraints while enabling automated discovery during generation. Evaluated on the MP-20 polymorph split, the method achieves a METRe accuracy of 79.06%, substantially outperforming the previous state-of-the-art baseline at 70.87%. Furthermore, it establishes new state-of-the-art matching performance on both the MP-20 and MPTS-52 standard CSP benchmarks.

Abstract

We introduce Human-AI Co-discovery system (HACO) for scientific algorithm discovery through cross-domain search and sparse human steering. Starting from the goal of generating crystal structures from chemical compositions, HACO searched across generative modeling methodologies from multiple fields and identified MaskGIT, a masked generative model from vision, as a promising framework for crystal structure prediction (CSP). HACO instantiated this masked formulation as a discrete token model of crystal structure; guided by sparse high-level human objectives, it then added crystallographic symmetry tokens, space group stratified sampling for polymorph coverage, and sub-bin coordinate refinement, yielding the Masked Generative Crystal Transformer (MaskGXT). On the MP-20 polymorph split, MaskGXT reaches 79.06% match-everyone-to-reference (METRe) accuracy, compared with 70.87% for the strongest evaluated baseline. MaskGXT also attains the best match rate on standard MP-20 and MPTS-52 CSP benchmarks. These results provide evidence that, in domains offering cheap, fast, and well-aligned validation, transfer-guided interactive AI co-scientists can contribute to scientific algorithm discovery by identifying transferable modeling principles and combining them with targeted human domain guidance.

Problem

• crystal structure prediction
• polymorph
• generative modeling
• materials discovery
• CSP

Innovation

• Masked Generative Model
• Crystal Structure Prediction
• Human-AI Co-discovery
• Cross-domain Transfer
• Symmetry-aware Tokenization